Recently, text-to-image generative Artificial Intelligence (AI) models have demonstrated their ability to generate high-quality art with text prompts. However, generative AI is still incapable of creating images that precisely reflect emotions. We propose Prompirit, an automatic prompt engineering assistance for improving AI-generated art in terms of expressiveness of emotion and aesthetics. We explored various approaches to refine users' free-form text inputs by incorporating user emotion and style keywords. Statistical analysis and user evaluation with 100 respondents showed that Prompirit significantly improved the alignments of image-emotion and the aesthetics of the generated image while precisely conveying the content of the original text input. Based on the results, we provide implications for creating affective images.

• Paper

• Supplymentary Material